Juxtaposition of different machine learning techniques for improved time series classification

Abstract

An essential type of TS analysis is classification, which can, for instance, advance energy load forecasting in smart grids by discovering the varieties of electronic gadgets based totally on their strength expenditure profiles recorded by way of computerized sensors. Such applications are very often characterised by using (a) very lengthy TS and (b) extensive TS datasets needing classification. but, current techniques to time series classification (TSC) cannot deal with such facts volumes at desirable accuracy. WEASEL (Word ExtrAction for time SEries cLassification), a novel TSC method which is each rapid and unique. Like different today's TSC techniques, WEASEL modifies time collection into characteristic vectors, the use of a sliding-window approach, which is then surpassed via a device getting to know classifier. Our approach here is the amalgamation of Distance-specific approaches such as DTW alongwith feature-specific approaches namely SAX and WEASEL and hence, this method may be effortlessly prolonged to be used in aggregate with different strategies. specially, we show that once blended with the space measures which include Minkowski distance measures, DTW, SAX and PAA, it outperforms the previously known methods.